T2S: Tokenized Skill Scaling for Lifelong Imitation Learning

TL;DR

T2S introduces a unified tokenized framework for lifelong imitation learning that effectively prevents forgetting, scales new skills efficiently, and transfers knowledge across tasks with minimal parameter growth.

Contribution

The paper proposes Tokenized Skill Scaling (T2S), a novel transformer-based approach that enhances scalability and knowledge transfer in lifelong imitation learning.

Findings

Prevents catastrophic forgetting with 1.0% average NBT.

Scales new skills with only 8.0% trainable tokens.

Achieves 77.7% average FWT for knowledge transfer.

Abstract

The main challenge in lifelong imitation learning lies in the balance between mitigating catastrophic forgetting of previous skills while maintaining sufficient capacity for acquiring new ones. However, current approaches typically address these aspects in isolation, overlooking their internal correlation in lifelong skill acquisition. We address this limitation with a unified framework named Tokenized Skill Scaling (T2S). Specifically, by tokenizing the model parameters, the linear parameter mapping of the traditional transformer is transformed into cross-attention between input and learnable tokens, thereby enhancing model scalability through the easy extension of new tokens. Additionally, we introduce language-guided skill scaling to transfer knowledge across tasks efficiently and avoid linearly growing parameters. Extensive experiments across diverse tasks demonstrate that T2S: 1) effectively prevents catastrophic forgetting (achieving an average NBT of 1.0% across the three LIBERO task suites), 2) excels in new skill scaling with minimal increases in trainable parameters (needing only 8.0% trainable tokens in an average of lifelong tasks), and 3) enables efficient knowledge transfer between tasks (achieving an average FWT of 77.7% across the three LIBERO task suites), offering a promising solution for lifelong imitation learning.